Cheap fakes
Genuine images or recordings altered with ordinary tools, slowed, sped up, cropped, spliced, or simply relabelled, so that they appear to show something they do not.
No machine learning is needed to deceive with a picture. A real photograph with a false caption, footage from one disaster presented as another, a speech clipped to reverse its meaning, a video slowed to make its subject seem impaired: each is cheap to make and, because the material is genuine, survives the tests that would catch a forgery. The image is authentic; the claim attached to it is not.
Paris and Donovan's "spectrum of audiovisual manipulation" placed cheap fakes at the accessible end and deepfakes at the expensive one, and observed that cheap fakes have done most of the actual damage, because they are fast, easy, and, being real footage, hard to rebut without the original context. The device is quoting out of context applied to media. Exploits Seeing Is Believing and Anchoring.
Historical examples
Paris and Donovan's report for Data & Society cataloged the techniques, recontextualising, speeding and slowing, lookalikes, face-swapping with consumer apps, and documented cases in which each had circulated as genuine, from doctored footage of public figures to misattributed war footage. They argued that the cheap end of the spectrum, not the expensive one, is where policy attention belonged.
Paris, Britt, and Joan Donovan. Deepfakes and Cheap Fakes: The Manipulation of Audio and Visual Evidence. New York: Data & Society Research Institute, 2019. Source
Every major storm since Hurricane Sandy in 2012 has been accompanied by the same recirculated images: a shark on a flooded highway, a dramatic cloud over a city that was not the one named, footage from a previous year's flood. Fact-checkers documented the same photographs reappearing for Harvey in 2017 and Florence in 2018 with new captions each time.
Snopes.com, "Hurricane Shark" fact-check archive, 2012–2018; Rosenblatt, Kalhan. "Fake Hurricane Harvey Photos Are Circulating on Social Media." NBC News, 28 August 2017.
In advertising and commerce
"Before and after" photographs taken in different lighting, posture, and time of day; a competitor's product filmed at a bad angle; a clipped quotation from a review video that inverted its verdict: the retail form of the cheap fake is routine enough that advertising codes address photographic "misleading impressions" specifically.
Advertising Standards Authority (UK). CAP Code, rule 3.1 on misleading advertising and guidance on the use of production techniques in cosmetics advertising, 2011.
Take a dramatic image from this week's news feeds and run a reverse image search. When did it first appear, and with what caption?
In the wild
Short examples of the technique at work, each with what to notice. The same examples are questions in the quiz, where the answer is not given away.
A video of the mayor "drunk at a council meeting" turns out to be the official recording slowed to 75 percent.
What to notice. Genuine footage altered with an ordinary tool; the realism comes from the material being real.
"Scenes from last night's riot," reads the caption on a photograph first published in another country in 2011.
What to notice. A real image with a false context; the relabelling is the whole forgery.
The clip shows the scientist saying "the vaccine is dangerous"; the full sentence was "the claim that the vaccine is dangerous has no support."
What to notice. Quoting out of context applied to video; the cut is the cheap fake.
Sources
- Paris, Britt, and Joan Donovan. Deepfakes and Cheap Fakes. New York: Data & Society Research Institute, 2019.
- Wardle, Claire, and Hossein Derakhshan. Information Disorder: Toward an Interdisciplinary Framework for Research and Policymaking. Strasbourg: Council of Europe, 2017.
- Silverman, Craig, ed. Verification Handbook. Maastricht: European Journalism Centre, 2014.